Control method and system for reducing power consumption after screen turn-off of mobile phone

Abstract

The invention provides a control method and system for reducing power consumption after screen turn-off of a mobile phone, and the method comprises the steps: obtaining daily behavior data of a user, and carrying out the preliminary analysis of the behavior data fusion time sequence; constructing a social behavior model and a shopping behavior model, respectively training, and predicting a behavior trend and a purchase intention of the user; dynamically setting the priority of the background task according to the prediction result; dynamically allocating resources in combination with priorities; a low-power-consumption preloading mechanism is designed, when a user turns off a screen, data is preloaded in advance to quickly access certain high-priority tasks, and the waiting time after the screen is lightened is shortened; in combination with changes of use habits of the users, prediction combination of the social behavior model and the shopping behavior model and a resource allocation strategy are optimized, and long-term requirements of the users are met. According to the invention, a plurality of defects in power consumption management in the prior art are overcome, and more intelligent and more efficient power consumption management in a screen turn-off state is realized especially in the aspect of dynamically adapting to user requirements.

Description

Technical Field

[0001] The present invention belongs to the field of mobile phone energy consumption control, and in particular relates to a control method and system for reducing power consumption of a mobile phone after the screen is turned off. Background Art

[0002] With the rapid popularization of mobile devices, smartphones have become an indispensable tool in daily life. Smartphones have become increasingly versatile, ranging from basic communications and social interactions to online shopping, media consumption, advertising push and other applications. Users have become more and more dependent on smartphones. However, with the increase in device functions and the diversification of application requirements, the problem of battery life has become increasingly prominent. Even when the smartphone is in the screen-off state, a large number of background tasks are still running continuously, especially applications that are resident in the background, such as social interaction, shopping recommendations, and advertising push. The continuous operation of these applications poses a considerable challenge to battery life. Traditional power management strategies mainly rely on background task optimization solutions provided by application developers or unified scheduling of operating systems, usually based on simple "fixed rules" to suspend or reduce background activities. However, this management method based on static rules often cannot dynamically adapt to users' personalized needs, and cannot effectively distinguish users' actual usage intentions and needs after the screen is off.

[0003] For example, a user may be engaging in an important social interaction or shopping decision before turning off the screen. The existing background process management system cannot intelligently identify these scenarios, which sometimes causes users to miss key message push or promotional information. When users do not need certain functions, such as social or shopping applications during inactive periods, the system fails to stop background processes in time, which further increases the waste of power. In addition, applications for personalized advertising push and shopping recommendations often make frequent network data requests when the screen is off, and even if these pushes are useless for a short time after the screen is off, they continue to consume power. This "one-size-fits-all" background management mechanism does not effectively combine user behavior patterns for dynamic adjustment, resulting in an obvious contradiction between meeting user needs and optimizing power consumption.

[0004] Although existing technologies attempt to optimize resource allocation through certain intelligent power management functions (such as identifying environmental conditions through sensors, freezing low-priority applications after the screen is off, etc.), most of these solutions remain at the simple "scene detection" level and are unable to deeply analyze user behavior patterns and personalized needs, resulting in the system's power consumption optimization in actual usage scenarios not being intelligent enough. Especially when applications are closely related to social and shopping behaviors, existing technologies fail to effectively handle such complex dynamic needs and are unable to achieve truly intelligent power consumption management when the screen is off. The complexity and diversity of user behavior make traditional power optimization methods unable to meet the needs of the intelligent era for precise and dynamic power management. Existing power management methods are too simple and lack flexibility, so there are still major deficiencies in meeting user experience and extending battery life. Summary of the invention

[0005] The purpose of the present invention is to propose a control method and system for reducing power consumption of a mobile phone after the screen is turned off. By intelligently analyzing the user's social and shopping behaviors, a personalized background task management and push control strategy are designed, which overcomes multiple defects of the prior art in power consumption management, especially in dynamically adapting to user needs, and realizes smarter and more efficient power consumption management in the screen-off state.

[0006] In order to achieve the above object, a first aspect of the present invention provides a control method for reducing power consumption of a mobile phone after the screen is turned off, the method comprising:

[0007] S1. Obtain users’ daily behavior data, including social interaction data and shopping behavior data, and conduct preliminary analysis on the behavior data fusion time series;

[0008] S2. Based on the results of the preliminary analysis, a social behavior model and a shopping behavior model are constructed, and they are trained separately to predict the user's behavior trend and purchase intention; wherein the social behavior model and the shopping behavior model use a recursive neural network structure for time series modeling;

[0009] Among them, the prediction of the social behavior model is:

[0010]

[0011] in, is the predicted value of social behavior at time t, indicating the predicted social activity; is the social behavior characteristics of the user at time t-1; It is a social behavior fluctuation adjustment item, which is used to handle the situation where users suddenly use social applications frequently or do not use them at all. is a short-term trend term representing social behavior, which is used to capture the rapid changes in user social behavior; α1 and β1 are weight coefficients, representing the impact of the adjustment term on the overall prediction result; Wsocial is the weight matrix obtained through model training; b social is the bias term, which controls the overall translation of the model; f is the activation function, and the nonlinear function ReLU or LeakyReLU is selected to increase the complexity of the prediction;

[0012] The predictions of the shopping behavior model are:

[0013]

[0014] in, is the predicted value of shopping behavior at time t; is the shopping behavior characteristics of the user at time t-1; It is a purchase intention adjustment item, which reflects the short-term changes in users' shopping interests. is the short-term trend term of shopping behavior; α2 and β2 are the corresponding weight parameters, controlling the contribution of the adjustment term; W shop is the weight matrix of the model, which is the core parameter of model learning; λ is the regularization parameter, which is used to limit the complexity of the model and prevent overfitting; g is the activation function, which is used to output nonlinear change results, and LeakyReLU is commonly used; ||W shop || 2 is a regular term used to prevent overfitting of the weight matrix;

[0015] S3. Dynamically set the priority of the background task according to the prediction result; wherein the priority is designed as follows:

[0016]

[0017] in, is the priority of the i-th background task, which determines the priority of its resource allocation; is the predicted value of social behavior related to the i-th task; is the predicted value of shopping behavior related to the i-th task; and are the social and shopping power consumption coefficients of the ith task, respectively, reflecting the impact of the task on battery consumption; α1 and α2 are the weight parameters of social and shopping behaviors, indicating the impact of the weights of different types of tasks in the priority calculation; U t is the current remaining power percentage of the system. When the power is low, the task priority will be more strictly regulated; E t is the current load level of the system, which is used to adaptively adjust the task priority; γ is the adjustment coefficient of system load and power, which is used to adjust the trade-off between power and load in different scenarios; It is a behavior fluctuation item, which is used to measure the volatility and suddenness of user behavior in order to adjust the priority;

[0018] Then introduce the resource allocation matrix to allocate specific system resources to background tasks according to their priorities, expressed as:

[0019]

[0020] in, is the resource allocation for the i-th task; R total is the total amount of system resources available; N is the total number of tasks currently running in the background

[0021] S4. Dynamically allocate resources based on priority to ensure that high-priority tasks receive sufficient resource support, while controlling resource usage of low-priority tasks and optimizing power consumption;

[0022] S5. Design a low-power preloading mechanism. When the user turns off the screen, preload data in advance to quickly access certain high-priority tasks, reducing the waiting time after the screen turns on.

[0023] S6. Based on the changes in user usage habits, optimize the prediction combination and resource allocation strategy of social behavior models and shopping behavior models to adapt to the long-term needs of users.

[0024] Preferably, the S1 further includes preprocessing the behavior data; the preliminary analysis of the behavior data fusion time series is specifically:

[0025] Perform feature mapping on the preprocessed data to quantify the activity of user behavior, including:

[0026] For social interaction data:

[0027]

[0028] in, represents the social activity of the user at time t, w i is the weight of different social data dimensions, u i is the collected user social behavior, n is the total number of social interaction users, and i is the social interaction user i;

[0029] For shopping behavior data:

[0030]

[0031] Among them, w j is the weight of each dimension of shopping behavior, v j is the user's shopping data, m is the total number of shopping users, and j is shopping user j;

[0032] The processed behavioral characteristics and Combined with the time series, it is represented and stored as follows: Where T is the total number of time segments collected, and The characteristic values ​​of social and shopping behaviors at each moment.

[0033] Preferably, the social behavior fluctuation adjustment item The definition is as follows:

[0034]

[0035] Among them, μ social is the mean of the social behavior characteristics; N is the size of the backtracking time window, which is used to determine the historical length of the evaluation;

[0036] The purchase intent adjustment The definition is as follows:

[0037]

[0038] in, is the user's recent shopping behavior data; c i is the weight of different behavior types, indicating the importance of different shopping behaviors; M is the size of the most recent time window.

[0039] Preferably, the behavior fluctuation term Fine-tune the priority based on the rate of change of the user's recent behavior, expressed as:

[0040]

[0041] in, is the behavior feature value of the i-th task at time t; M is the length of the retrospective time window, which is used to evaluate the historical period of behavior fluctuations;

[0042] The social and shopping power consumption coefficients of the i-th task and Among them, the power consumption coefficient of social tasks Reflects the power consumption of the i-th social task and the power consumption coefficient of the shopping task Indicates the power consumption of shopping-related tasks. Dynamically adjust according to its active and passive consumption behavior; wherein the power consumption coefficient of the social task The dynamic adjustment is expressed as:

[0043]

[0044] in represents the active consumption coefficient of the i-th task, that is, the power consumption when the task is active; Indicates the passive consumption coefficient of the task, such as the power consumption during data synchronization; ω social A parameter used to adjust the weight between active and passive power consumption.

[0045] Preferably, the objective optimization function of S4 is:

[0046]

[0047] in, is the amount of resources allocated to the i-th task; is the priority of the ith task, reflecting the resource urgency of the task in the current system; N is the total number of current background tasks; R total is the total resources available in the current system; R battery U is the total amount of resources adjusted based on the battery power, indicating the amount of resources that the system can allocate when the battery power is low; t The current system battery percentage; E t is the current load status of the system, indicating the overall busyness of the system; η is the weight coefficient used to balance the total amount of resources R total and the resource R adjusted based on the battery condition battery ;

[0048] At the same time, the upper limit of task power consumption is designed, and the power consumption threshold of the task is dynamically set according to the battery status and system load, which is expressed as:

[0049]

[0050] in, is the upper limit of power consumption of the i-th task, indicating the maximum power consumption that the task can consume; P max is the maximum power consumption limit of the task; U t The current battery level of the system; E t is the current load status of the system, indicating whether the system is running under high load; power It is the power consumption adjustment factor, which determines the adjustment range of power consumption under different battery and load conditions. When the battery condition is good, the task obtains resources close to the maximum power consumption, and when the battery power is the lowest or the system load is the highest, the power consumption limit will be reduced to the minimum, ensuring that unnecessary tasks do not waste power.

[0051] Preferably, the S4 also includes designing a task freezing mechanism, when the system is in an extreme situation, freezing resources for low-priority tasks and releasing more resources for high-priority tasks, which is expressed as:

[0052]

[0053] in, is the freezing state of the i-th task, 1 means freezing, 0 means not freezing; θ freeze is the priority threshold for task freezing. Tasks below this threshold will be frozen when the battery is low. min The minimum power threshold set for the system. When the power is lower than this threshold, some low-priority tasks need to be frozen.

[0054] Preferably, the S5 further includes designing an adaptive adjustment factor, adjusting the factor based on the historical load and power state of the system, and continuously optimizing the resource and power consumption management strategy, which is expressed as:

[0055]

[0056] Among them, λ adaptive is a dynamic adjustment factor used to adjust the sensitivity of resource and power management; initial is the initial setting of the adjustment factor value; δ is the adjustment increment coefficient, which controls the adjustment range of the adjustment factor; U t is the battery charge at time t; E t is the system load status at time t; T is the system historical data at the past T times.

[0057] Preferably, the S5 specifically includes:

[0058] Design a low-power preloading mechanism. When the device enters the screen-off mode, although the user will not use the device immediately, the background task still needs to process key operations, which can be expressed as:

[0059]

[0060]

[0061] in, is the preloading coefficient of the i-th task, indicating the preloading amount allowed for the task in low power mode; is the priority of the ith task, which comes from the previous step and is calculated based on the social and shopping behavior prediction and power consumption model; is the resource allocation of the ith task, reflecting the proportion of the task in the system resources; The maximum preloading power consumption threshold set for the system is used to limit the total power consumption of the preloading task to ensure that the system does not consume excessive power due to preloading; is the scheduling coefficient of the i-th task, indicating the resource scheduling priority of the task in the screen-off state; is the preloading coefficient of the i-th task, reflecting the preloading requirement of the task; is the resource allocation ratio of the i-th task; N is the total number of active background tasks in the current system; θ is the balance coefficient, which controls the scheduling ratio of preloaded tasks and actual running tasks; λ preload is the power consumption-preload adjustment factor, which controls the power consumption sensitivity of the control system to the preload task in the low power state; λ base is the initial preload power consumption sensitivity; α is the adjustment increment, which reflects the power consumption control strength of the system on the preload task according to the load and power conditions; U t The remaining battery percentage of the current system; E t It is the current system load status, indicating the overall busyness of the system.

[0062] Preferably, the S6 specifically includes:

[0063] Design an adaptive learning mechanism to learn from the actual resource consumption and power consumption data of tasks through feedback loops, and dynamically optimize task scheduling and power consumption management strategies.

[0064] In another aspect of the present invention, a control system for reducing power consumption of a mobile phone after the screen is turned off is provided, the system comprising:

[0065] Daily behavior data acquisition unit, used to acquire the user's daily behavior data, including social interaction data and shopping behavior data, and conduct preliminary analysis on the behavior data fusion time series;

[0066] A model building unit, used to build a social behavior model and a shopping behavior model based on the results of the preliminary analysis, and to train them respectively to predict the user's behavior trend and purchase intention; wherein the social behavior model and the shopping behavior model use a recursive neural network structure for time series modeling;

[0067] Among them, the prediction of the social behavior model is:

[0068]

[0069] in, is the predicted value of social behavior at time t, indicating the predicted social activity; is the social behavior characteristics of the user at time t-1; It is a social behavior fluctuation adjustment item, which is used to handle the situation where users suddenly use social applications frequently or do not use them at all. is a short-term trend term representing social behavior, which is used to capture the rapid changes in user social behavior; α1 and β1 are weight coefficients, representing the impact of the adjustment term on the overall prediction result; W social is the weight matrix obtained through model training; b socialis the bias term, which controls the overall translation of the model; f is the activation function, and the nonlinear function ReLU or LeakyReLU is selected to increase the complexity of the prediction;

[0070] The predictions of the shopping behavior model are:

[0071]

[0072] in, is the predicted value of shopping behavior at time t; is the shopping behavior characteristics of the user at time t-1; It is a purchase intention adjustment item, which reflects the short-term changes in users' shopping interests. is the short-term trend term of shopping behavior; α2 and β2 are the corresponding weight parameters, controlling the contribution of the adjustment term; W shop is the weight matrix of the model, which is the core parameter of model learning; λ is the regularization parameter, which is used to limit the complexity of the model and prevent overfitting; g is the activation function, which is used to output nonlinear change results, and LeakyReLU is commonly used; ||W shop || 2 is a regular term used to prevent overfitting of the weight matrix;

[0073] The priority design unit is used to dynamically set the priority of the background task according to the prediction result; wherein the priority design is as follows:

[0074]

[0075] in, is the priority of the i-th background task, which determines the priority of its resource allocation; is the predicted value of social behavior related to the i-th task; is the predicted value of shopping behavior related to the i-th task; and are the social and shopping power consumption coefficients of the ith task, respectively, reflecting the impact of the task on battery consumption; α1 and α2 are the weight parameters of social and shopping behaviors, indicating the impact of the weights of different types of tasks in the priority calculation; U t is the current remaining power percentage of the system. When the power is low, the task priority will be more strictly regulated; E t is the current load level of the system, which is used to adaptively adjust the task priority; γ is the adjustment coefficient of system load and power, which is used to adjust the trade-off between power and load in different scenarios; It is a behavior fluctuation item, which is used to measure the volatility and suddenness of user behavior in order to adjust the priority;

[0076] Then introduce the resource allocation matrix to allocate specific system resources to background tasks according to their priorities, expressed as:

[0077]

[0078] in, is the resource allocation for the i-th task; R total is the total amount of system resources available; N is the total number of tasks currently running in the background

[0079] Resource allocation unit, used to dynamically allocate resources according to priority, ensuring that high-priority tasks receive sufficient resource support, while controlling resource occupancy of low-priority tasks and optimizing power consumption;

[0080] The low-energy loading unit is used to design a low-power preloading mechanism. When the user turns off the screen, the preloading data can be used to quickly access certain high-priority tasks, thus reducing the waiting time after the screen turns on.

[0081] The system optimization unit is used to optimize the prediction combination and resource allocation strategies of the social behavior model and the shopping behavior model based on changes in user usage habits to adapt to the long-term needs of users.

[0082] The beneficial technical effects of the present invention are at least as follows:

[0083] In view of the defects of the prior art in power consumption management when the screen is off, the present invention proposes an intelligent screen-off power consumption management system based on social behavior and shopping behavior analysis. The system conducts in-depth analysis of users' social interactions and e-shopping behaviors, combines users' personalized usage patterns, dynamically adjusts the priority and resource allocation of background tasks, optimizes power consumption management, and achieves a balance between user needs and battery life.

[0084] First, the present invention introduces a social behavior analysis module, which collects the user's social activity data before and after the screen is turned off in real time to intelligently determine the user's social needs. If the user is frequently interacting with certain specific contacts before the screen is turned off, or has unread important messages, the module will dynamically manage the background process of the social application after the screen is turned off to ensure that the user will not miss important message pushes; and for social behaviors with lower activity, the system will automatically reduce the background priority of related applications to avoid unnecessary resource occupation and power consumption. This module overcomes the deficiency of the existing technology that it cannot effectively distinguish between user interaction intensity and social needs, and realizes personalized power consumption optimization for social behaviors.

[0085] Secondly, the present invention also includes a shopping behavior analysis module, which intelligently predicts the user's shopping needs when the screen is off based on the user's browsing history, collection preferences and purchase history. For users with potential shopping behavior, the system will continue to keep the shopping push active to ensure that important promotional information and product recommendations can be pushed to the user in a timely manner; and for situations where shopping needs are not obvious, the system will delay or stop the running of the background shopping application to reduce unnecessary network data requests and resource usage, thereby effectively reducing battery consumption. This function solves the problem in the prior art that shopping needs cannot be intelligently judged and power consumption is wasted due to blindly pushing information.

[0086] The third important innovation of the present invention is the personalized power consumption optimization engine, which combines the user's social behavior and shopping behavior, adopts an adaptive learning mechanism, and dynamically adjusts the resource allocation and network request frequency of background applications through in-depth learning of the user's long-term usage pattern. The system can continuously optimize the power consumption management strategy based on the user's historical usage data, and gradually improve the ability to respond to the user's personalized needs. This engine overcomes the shortcomings of the rigid power consumption management strategy and lack of personalized adjustment in the prior art, allowing the system to intelligently adapt to the user's different usage scenarios and further improve battery life.

[0087] In summary, the present invention designs personalized background task management and push control strategies by intelligently analyzing users' social and shopping behaviors, thereby overcoming multiple defects of the prior art in power consumption management, especially in dynamically adapting to user needs, and achieving smarter and more efficient power consumption management in the screen-off state. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0089] Figure 1 This is a flow chart of a control method for reducing power consumption of a mobile phone after the screen is turned off according to an embodiment of the present invention.

[0090] Figure 2 This is a framework diagram of a control system for reducing power consumption of a mobile phone after the screen is turned off according to an embodiment of the present invention. DETAILED DESCRIPTION

[0091] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0092] like Figure 1 As shown, an embodiment of the present invention provides a control method for reducing power consumption of a mobile phone after the screen is off, and the method includes the following steps S1-S6:

[0093] S1. Obtain users’ daily behavior data, including social interaction data and shopping behavior data, and conduct preliminary analysis on the behavior data fusion time series;

[0094] Specifically, the user behavior dataset will contain two categories of data: social data and shopping data, denoted as D raw Among them, D raw It is the original behavior data set. The initial input raw data is obtained from sources such as mobile phone sensors, application logs, network interaction records, etc. This data will be used for subsequent model training and behavior prediction.

[0095] D raw The structure is as follows:

[0096] It is the social behavior data vector collected at time t, including multiple dimensions such as user operation behavior in social applications, number of message interactions, unread message status, usage time, etc.

[0097] is the shopping behavior data vector collected at time t, including the user's browsing on the e-commerce platform, shopping cart items, favorite records, advertising click data, etc.

[0098] Specific data example: where u n Data of the user's nth social application, such as the number of message interactions and the number of unread messages for each application.

[0099] Similarly, where v m Indicates the user's actions on an e-shopping platform, such as the number of clicks, the number of product categories browsed, etc.

[0100] Collecting data and The dataset D that constitutes the initial input raw , for subsequent analysis. The time dimension of this data is necessary because behavioral analysis relies on temporal trends.

[0101] Furthermore, the original behavior data D raw It often contains noisy or invalid data, such as incomplete log records, blank sensor inputs, etc., which can mislead subsequent models. Therefore, the present invention requires a unique pre-processing step for this inventive scenario to remove noise and correct the data.

[0102] Further, the preprocessing steps:

[0103] The present invention ensures that the collected social data is With shopping data The timestamps are aligned, which ensures that all dimension data are successfully collected at the same time.

[0104] When there are missing data in some time segments (such as or Some values ​​in are empty), and linear interpolation technology is used to fill them. Calculate using data from adjacent moments:

[0105]

[0106] in, is the completed data at time t, S t-1 With S t+1 are the data before and after respectively.

[0107] Furthermore, the original data is often too high-dimensional and contains redundant information, so it is necessary to extract features that can effectively represent user behavior and reduce the dimensionality of the data. The steps are:

[0108] A behavior feature mapping method based on frequency and intensity is designed to quantify the activity of user behavior. For example, the frequency of use of social applications and the intensity of message interaction can be obtained by the following formula:

[0109]

[0110] in, represents the social activity of the user at time t, w i is the weight of different social data dimensions (e.g., the importance of unread message status is higher than browsing time), u i The collected user social behaviors.

[0111] Correspondingly, the user's shopping behavior activity is:

[0112]

[0113] where w j is the weight of each dimension of shopping behavior (such as the number of views and the importance of product clicks), v j The user's shopping data.

[0114] Furthermore, feature selection techniques are applied to transform high-dimensional data into low-dimensional behavioral feature representations. and This method can be used to convert the original high-dimensional dataset and Compressed into a pair of time series, representing the user's social activity and shopping behavior activity, thereby simplifying the subsequent analysis and modeling process.

[0115] Furthermore, the behavioral characteristics after processing and It needs to be stored in the time dimension so as to be used for subsequent model training and prediction. Through time serialization, the present invention can better capture the changing trend of user behavior over time, and then provide input for the prediction model of social and shopping behavior.

[0116] Furthermore, the present invention represents the above two behavior characteristics as a time series:

[0117]

[0118] Where T is the total number of time segments collected, and The characteristic values ​​of social and shopping behaviors at each moment.

[0119] Furthermore, time series data and It is stored in a structured database for model training and behavior prediction. This time series not only retains the dynamic changes of user behavior, but also simplifies the subsequent analysis process so that the behavior characteristics at each time point can be tracked.

[0120] S2. Based on the results of the preliminary analysis, a social behavior model and a shopping behavior model are constructed and trained separately to predict the user's behavior trends and purchase intentions; wherein the social behavior model and the shopping behavior model use a recursive neural network structure for time series modeling.

[0121] Specifically, in the prediction of social and shopping behaviors, the present invention designs a user behavior dynamic prediction model (UBDP), which combines two information sources: 1) historical behavior data of users; 2) short-term fluctuation trends of user behaviors. In order to capture these dynamic behaviors, the model adds adaptive regularization terms and weight adjustment terms related to user behavior characteristics on the basis of traditional recurrent neural networks to meet the specific needs of the present invention.

[0122] Model structure: M social and M shop The recursive neural network (RNN) structure is used for time series modeling to predict future behavioral needs. At the same time, behavioral stability regularization terms and personalized adjustment terms are introduced into the loss function to dynamically optimize the characteristics of user behavior.

[0123] Furthermore, the social behavior model (M) of the present invention social) It is necessary to predict the user's future social activity. Based on the autoregressive mechanism of recurrent neural network (RNN), it can capture the user's long-term behavioral dependence. In order to adapt to the randomness and volatility of social behavior, the present invention innovatively introduces a social behavior fluctuation adjustment item to smooth the frequency of abnormal behavior of users. Formula definition:

[0124]

[0125] in, represents the predicted value of social behavior at time t, indicating the predicted social activity; Represents the social behavior characteristics of the user at time t-1; Represents the social behavior fluctuation adjustment item, which is used to handle the situation where users suddenly use social applications frequently or do not use them; represents the short-term trend of social behavior, which is used to capture the rapid changes of user social behavior; α1 and β1 are weight coefficients, representing the impact of the adjustment item on the overall prediction result; W social represents the weight matrix, obtained through model training; b social represents the bias term, which controls the overall translation of the model; f represents the activation function, and the nonlinear function ReLU or LeakyReLU is selected to increase the complexity of the prediction.

[0126] Furthermore, this item is an innovative part of the model, which is used to deal with the situation where the user's social behavior suddenly increases or decreases significantly within a certain period of time. The present invention defines it based on the variance change of user behavior:

[0127]

[0128] Among them, μ social It is expressed as the mean of social behavior characteristics; N represents the size of the time window for looking back, which is used to determine the historical length of the evaluation. This adjustment item can prevent the model from being disturbed by excessively frequent or infrequent social behaviors through historical fluctuation information, and maintain a reasonable prediction of the user's overall social behavior.

[0129] Furthermore, the core of shopping behavior prediction is to capture the user's short-term purchasing desire and long-term shopping trend. The present invention introduces a user purchase intention adjustment item, which is corrected based on the user's short-term shopping behavior changes, and combines regularization to limit the model complexity. The model adopts a recursive neural network architecture and introduces an innovative weight adjustment mechanism for shopping behavior. Formula definition:

[0130]

[0131] in, represents the predicted value of shopping behavior at time t; Represents the shopping behavior characteristics of the user at time t-1; Indicates the purchase intention adjustment item, reflecting the short-term changes in users' shopping interests; It is represented as the short-term trend term of shopping behavior; α2 and β2 are represented as the corresponding weight parameters to control the contribution of the adjustment term; W shop represents the weight matrix of the model, which is the core parameter of model learning; λ represents the regularization parameter, which is used to limit the complexity of the model and prevent overfitting; g represents the activation function, which is used to output nonlinear change results, and LeakyReLU is commonly used; ||W shop || 2 Represents a regular term, which is used to prevent overfitting of the weight matrix.

[0132] Furthermore, the adjustment item is defined based on the frequency of the user's recent shopping behavior, and the specific calculation formula is:

[0133]

[0134] in, Represented as the user's recent shopping behavior data (such as number of clicks, browsing time); c i It is represented by the weight of different behavior types, indicating the importance of different shopping behaviors; M represents the size of the most recent time window. Through this adjustment item, the model can capture the user's short-term shopping intention and identify the user's sudden purchase interest, thereby improving the accuracy of shopping behavior prediction.

[0135] Further, the output result is: P social represents the predicted result of the user's future social behavior; P shop Represents the predicted result of the user's future shopping behavior. These two results are the core inputs for setting the priority of background tasks in the subsequent steps. The social and shopping behaviors predicted by the model can help the system dynamically adjust task priorities, thereby more intelligently managing power consumption after the screen is off.

[0136] It can be understood that the present invention introduces the social behavior fluctuation adjustment item Smoothing out the frequency of abnormal user behavior improves the robustness of social behavior prediction. Designing purchase intention adjustment items It can sensitively capture the fluctuations of users' short-term shopping interests and enhance the accuracy of shopping behavior prediction. The regular term λ is used in the shopping behavior model to prevent the model from overfitting. At the same time, it takes into account the combination of behavioral trends and purchase intentions to optimize the performance of the model in shopping scenarios.

[0137] S3. Dynamically set the priority of background tasks according to the prediction results;

[0138] Specifically, the priority calculation model proposed in the present invention aims to dynamically assign task priorities based on the predicted value of user behavior, and introduces some innovative items in the scenario of the present invention to enhance the adaptability of the model to special scenarios. This includes: power consumption coefficient according to task type, behavior fluctuation items, and adaptive adjustment items of system status.

[0139] Priority calculation formula:

[0140]

[0141] in, Indicates the priority of the i-th background task, which determines the priority of its resource allocation; represents the predicted value of social behavior related to the i-th task, from M social Model; represents the predicted value of shopping behavior related to the i-th task, from M shop Model; and represents the social and shopping power consumption coefficient of the i-th task, reflecting the impact of the task on battery consumption; α1 and α2 represent the weight parameters of social and shopping behaviors, indicating the impact of the weights of different types of tasks in the priority calculation; U t Indicates the current remaining power percentage of the system. When the power is low, the task priority will be more strictly regulated; E t Indicates the current load level of the system (total number of tasks and system resource usage), which is used to adaptively adjust task priorities; γ represents the adjustment coefficient of system load and power, which is used to adjust the trade-off between power and load in different scenarios; The behavioral fluctuation term representing the innovative design is used to measure the volatility and suddenness of user behavior to adjust the priority.

[0142] Among them, the behavioral fluctuation term Indicates that this item fine-tunes the priority based on the rate of change of the user's recent behavior. For example, if the user's recent behavior changes frequently (such as suddenly becoming active on a social platform), such tasks should be given a higher priority to ensure that key interactions are not missed. Behavior fluctuation term formula:

[0143]

[0144] in, represents the behavioral feature value of the i-th task at time t (such as social activity or shopping interest); M represents the length of the retrospective time window, which is used to evaluate the historical period of behavioral fluctuations. This behavioral fluctuation term allows the system to make smarter priority adjustments based on the suddenness of user behavior. For example, a sudden surge in user social interactions should increase the priority of related tasks, while a stable behavior pattern will keep the priority unchanged.

[0145] Furthermore, when designing the priority of tasks, we should not only rely on the prediction of user behavior, but also consider the impact of the task itself on system power consumption. Each background task has its corresponding power consumption coefficient, which reflects its power consumption. Specifically:

[0146] Social task power consumption coefficient Reflects the power consumption of the ith social task. For example, a task that continuously keeps messages in sync consumes more power than a task that occasionally checks for new messages. Indicates the power consumption of shopping-related tasks, such as the power consumption of background operations such as ad push and data synchronization.

[0147] Dynamic adjustment of power consumption coefficient:

[0148]

[0149] in, represents the active consumption coefficient of the i-th task, that is, the power consumption when the task is active; Indicates the passive consumption coefficient of the task, such as the power consumption during data synchronization; ω social Represents the weight parameter used to adjust the active and passive power consumption, depending on the task type.

[0150] Similarly, for shopping tasks, It also dynamically adjusts based on its active and passive consumption behaviors. This approach ensures that priority setting is not only based on user needs, but also balanced according to the actual power consumption of the task, avoiding high-power consumption tasks from occupying resources unnecessarily.

[0151] Furthermore, the task priority is calculated Finally, the present invention introduces the resource allocation matrix R allocation , used to allocate specific system resources (such as CPU, memory, network bandwidth) to background tasks based on priority. In order to dynamically adapt to the resource requirements of different tasks, resource allocation not only depends on priority, but also needs to balance the current load of the system. Resource allocation formula:

[0152]

[0153] in, Indicates the resource allocation amount of the i-th task; represents the priority of the i-th task, calculated by the above formula; R total Indicates the total amount of system available resources (such as allocatable CPU, network bandwidth, etc.); N indicates the total number of tasks currently running in the background.

[0154] As can be understood, this resource allocation formula ensures that system resources are allocated in proportion to the priority of the tasks. Tasks with higher priorities receive a larger share of resources, while tasks with lower priorities are restricted in their resource usage, or even frozen when the power or system load limit is reached.

[0155] Furthermore, the priority setting scheme of the present invention allows the system to make adaptive adjustments based on fluctuations in user behavior, dynamic changes in power consumption coefficients, and the current state of the system (such as load and power). This mechanism ensures that background tasks can run efficiently in the screen-off state without causing unnecessary burden on battery life.

[0156] Furthermore, the task freezing mechanism: when Too low (lower than a certain set threshold θ low ), the system will directly freeze the task and stop all its activities until the system status returns to an acceptable level. Freezing tasks can reduce useless power consumption and ensure the normal operation of core tasks under high load or low power conditions.

[0157] Among them, this priority setting scheme uses the behavior prediction value P social and P shop , power consumption coefficient adjustment and and behavioral fluctuations Innovative mechanisms such as , etc., have achieved the goal of intelligent background task priority allocation. Compared with traditional power management methods, this solution can dynamically adapt to changes in user behavior and perform refined resource management based on task power consumption, system load and battery status.

[0158] The priority calculation combines the adjustment of user behavior fluctuations and the active and passive power consumption coefficients of the task to ensure that background tasks can remain active according to the user's actual needs when the screen is off, avoiding unnecessary power consumption, and greatly improving the system's responsiveness and battery life performance in complex scenarios.

[0159] S4. Dynamically allocate resources based on priority to ensure that high-priority tasks receive sufficient resource support, while controlling resource usage of low-priority tasks and optimizing power consumption;

[0160] Specifically, traditional resource allocation is usually based on task priority and total system resources. In this invention, not only does the invention need to dynamically adjust resource allocation based on priority, but it also needs to fine-grainedly control resources based on the current battery power status and system load to ensure that power consumption is minimized. To this end, a new set of resource allocation and power consumption management strategies are designed. Core formula: Resource allocation formula

[0161]

[0162] in, Represents the amount of resources (such as CPU time, memory, network bandwidth, etc.) allocated to the i-th task. Indicates the priority of the ith task, reflecting the resource urgency of the task in the current system. N indicates the total number of current background tasks. total Indicates the total amount of resources available in the current system (such as CPU, bandwidth). battery Indicates the total amount of resources adjusted based on the battery power, indicating the amount of resources that the system can allocate when the battery power is low. t Indicates the current system battery percentage. t Indicates the current load status of the system and the overall busyness of the system. η represents the weight coefficient, which is used to balance the total amount of resources R total and the resource R adjusted based on the battery condition battery .

[0163] The formula calculates resource allocation in two parts: one part comes from the weighted calculation of task priority, and the other part determines the resource allocation ratio based on the system battery power status and current load conditions. The weight η controls the balance between system resources and battery conditions. If the battery power is low, the battery adjustment part of the formula will play a greater role, and the resources allocated to the task will be more stringent, reducing unnecessary resource waste.

[0164] It is understandable that traditional resource allocation often depends on the priority of tasks, but in the field of power consumption management, the present invention must perform more detailed dynamic control of power. This formula introduces the remaining battery capacity U t and load regulation factor E t , effectively realizing dynamic resource regulation based on the real-time status of the system.

[0165] Furthermore, after determining resource allocation, the system needs to strictly control the power consumption of each task. The present invention designs a task power consumption upper limit P limit , dynamically set the power consumption threshold of the task according to the battery status and system load. Core formula: Power consumption management formula

[0166]

[0167] in, is the upper limit of the power consumption of the i-th task, indicating the maximum power consumption that the task can consume. max Indicates the maximum power consumption limit of the task, usually the power consumption limit set by the system. t Indicates the current battery level of the system. t Indicates the current load status of the system, indicating whether the system is running under high load. Indicates the priority of the task, which determines the importance of the task in power consumption allocation. power Indicates the power consumption adjustment factor, which determines the adjustment range of power consumption under different battery and load conditions.

[0168] Understandably, this formula sets the upper limit of the power consumption of the task. When the battery is in good condition, the task can obtain resources close to the maximum power consumption, and when the battery is low or the system load is high, the power consumption limit will be significantly reduced to ensure that unnecessary tasks do not waste power. This is a further control of system resources, which can effectively reduce the negative impact of background tasks on battery life.

[0169] Furthermore, an innovative task freezing mechanism is designed: when the system is in extreme conditions (such as extremely low power or excessive system load), the present invention designs an automatic task freezing mechanism to freeze resources for low-priority tasks and release more resources for high-priority tasks.

[0170] Task Freeze Formula

[0171]

[0172] in, Indicates the freezing state of the i-th task, 1 means freezing, and 0 means not freezing. θ freeze Indicates the priority threshold of task freezing. Tasks below this threshold will be frozen when the battery is low. min Indicates the minimum power threshold set by the system. When the power is lower than this level, some low-priority tasks need to be frozen.

[0173] Furthermore, when the system power is lower than the set threshold U min , and the task priority is lower than θ freeze When the task is called, the system will freeze the task and stop all its background activities. This mechanism can not only effectively release resources, but also ensure that the battery power is maximized for more important tasks, extending the device's battery life.

[0174] Furthermore, a dynamic adjustment mechanism is designed: In order to better adapt to the dynamic changes in user needs, the system also needs to continuously optimize resource and power consumption management strategies. The present invention introduces an adaptive adjustment factor λ adaptive , which is adjusted based on the system's historical load and power status to ensure that the system can adapt to the user's personalized behavior and habits in long-term use. Adaptive adjustment formula λ adaptive :

[0175]

[0176] Among them, λ adaptive Represents a dynamic adjustment factor used to adjust the sensitivity of resource and power management. initial Indicates the initial setting of the adjustment factor value. δ indicates the adjustment increment coefficient, which controls the adjustment range of the adjustment factor. t Indicates the battery charge at time t. E t Indicates the system load status at time t. T indicates the system historical data at the past T time points.

[0177] Furthermore, the adaptive adjustment factor adjusts the system's sensitivity to resources and power consumption based on historical battery and load conditions. For example, when the system load is low and the battery is sufficient, the adjustment factor will cause the system to relax the power consumption restrictions and allocate more resources to the task; when the system load is too high or the battery is insufficient, the adjustment factor will tighten the resource allocation and further reduce unnecessary power consumption.

[0178] Furthermore, this step effectively achieves the goal of the present invention through the combination of resource scheduling, power consumption management and task freezing mechanism: dynamically optimizing resource allocation in the screen-off state, reducing ineffective power consumption, and improving the system's long-term adaptability to user behavior through adaptive adjustment factors.

[0179] It is understandable that, unlike traditional power optimization, this solution comprehensively regulates multiple factors such as battery status, system load, and task priority, so that the system can still maintain the normal operation of key tasks when resources are tight, avoiding the problem of low-priority tasks occupying limited resources. In particular, the introduction of task freezing mechanism and dynamic power consumption upper limit setting can effectively optimize battery life in complex usage scenarios.

[0180] S5. Design a low-power preloading mechanism. When the user turns off the screen, preload data in advance to quickly access certain high-priority tasks, reducing the waiting time after the screen turns on.

[0181] Specifically, when the device enters the screen-off mode, the user may not use the device immediately, but background tasks (such as social messages, shopping recommendations, etc.) still need to process some key operations. In order to ensure the user experience (such as receiving important notifications in time after the screen is off), the present invention needs to preload these high-priority tasks in advance with low power consumption, and control the running behavior of these tasks in the low-power state. Preload formula

[0182]

[0183] in, It represents the preloading coefficient of the i-th task, which indicates the preloading amount allowed for the task in low power mode. It represents the priority of the ith task, which comes from the previous step and is calculated based on the social and shopping behavior prediction and power consumption model. Indicates the resource allocation of the i-th task, reflecting the proportion of the task in the system resources. Indicates the upper limit of the power consumption of the task to prevent the task from consuming too much power during preloading. Indicates the maximum preloading power consumption threshold set by the system, which is used to limit the total power consumption of the preloading task to ensure that the system does not consume excessive power due to preloading.

[0184] Among them, the task scheduling formula

[0185]

[0186] in, Represents the scheduling coefficient of the i-th task, which indicates the resource scheduling priority of the task in the screen-off state. It represents the preloading coefficient of the i-th task, reflecting the preloading requirement of the task. represents the resource allocation ratio of the i-th task. N represents the total number of active background tasks in the current system. θ represents the balance coefficient, which controls the scheduling ratio of preloaded tasks and actual running tasks.

[0187] Among them, the adjustment formula λ preload :

[0188]

[0189] Among them, λ preload Represents the power consumption-preload adjustment factor, which controls the power consumption sensitivity of the system to the preload task in the low power state. base Indicates the initial setting of preload power consumption sensitivity. α indicates the adjustment increment, which reflects the power consumption control strength of the system on the preload task according to the load and power conditions. U t Indicates the current system battery remaining percentage.t Indicates the current system load status and the overall busyness of the system.

[0190] S6. Based on the changes in user usage habits, optimize the prediction combination and resource allocation strategy of social behavior models and shopping behavior models to adapt to the long-term needs of users.

[0191] Specifically, the user's behavior pattern, task power consumption requirements and resource load are constantly changing, so the task scheduling and preloading strategy cannot remain unchanged. This step introduces an adaptive learning mechanism that learns from the actual resource consumption and power consumption data of the task through a feedback loop to dynamically optimize the task scheduling and power management strategy.

[0192] Furthermore, the present invention introduces a learning rate η learn , which is used to control how the system adjusts the scheduling and resource allocation strategy of future tasks based on historical behavior data. Learning rate formula η learn :

[0193]

[0194] in, Represents the adaptive learning rate of the i-th task, controlling the step size of learning optimization. η base Represents the basic learning rate and sets the initial sensitivity of learning. Indicates the actual amount of resources consumed by the i-th task. Indicates the amount of resources allocated to the i-th task. Indicates the actual power consumption of the i-th task. represents the upper limit of power consumption of the i-th task. α represents the resource difference weight, which controls the impact of resource difference on learning rate. β represents the power consumption difference weight, which controls the impact of power consumption difference on learning rate.

[0195] Among them, the adaptive learning rate The core of the algorithm is to dynamically adjust the learning rate based on the difference between the actual resource and power consumption of each task and the assigned value. When the actual resource consumption and power consumption of the task differ greatly from the preset value, the learning rate will increase, allowing the system to adjust the allocation strategy more quickly. On the contrary, if the difference is small, the learning rate will gradually decrease, reducing the frequency of the system's adjustment of task scheduling.

[0196] Furthermore, after collecting the resource and power consumption data of historical tasks, the system needs to adjust future task scheduling and power consumption management strategies through an adaptive learning mechanism to make the system better adapt to the dynamic needs of users.

[0197]

[0198] in, It represents the optimization scheduling coefficient of the i-th task and the resource scheduling weight of the task after learning. represents the initial scheduling coefficient of the i-th task, which is calculated in step 5. Represents the learning rate of the i-th task, which controls the optimization range of task scheduling.

[0199] Understandably, the optimized scheduling coefficient According to the learning rate Dynamic adjustment. By increasing the learning rate, the system can quickly respond to the actual needs of the task and improve the accuracy of resource allocation; when the actual performance of the task is close to the preset, the adjustment of the scheduling coefficient will become more stable. This method ensures that the system can optimize scheduling in low-power mode and gradually adapt to the user's personalized usage habits over time.

[0200] Furthermore, the adaptive learning system needs to continuously obtain the actual execution data of the task and adjust the task scheduling and preloading strategy through the feedback loop. The specific feedback process includes:

[0201] Actual resource consumption feedback: actual resource consumption data collected during task execution Used to update resource allocation policies.

[0202] Actual power consumption feedback: collects actual power consumption data of tasks Used to adjust power consumption limits.

[0203] Task execution result evaluation: Optimize the preloading strategy for future tasks by evaluating the actual execution results of the tasks (such as response speed, loading delay, etc.).

[0204] Feedback formula: Resource and power optimization update:

[0205]

[0206] in, Represents the updated resource allocation value of the i-th task. Represents the updated power consumption upper limit of the i-th task. It represents the learning rate of the i-th task, which determines the magnitude of resource and power consumption adjustments.

[0207] It is understandable that through the adaptive learning process, the resource allocation value and power consumption limit of the task are constantly updated, and the system can gradually optimize the resource occupancy and power consumption performance of each task. The feedback mechanism ensures that each task can get more appropriate resource allocation and power consumption limit in subsequent operation, ensuring the resource optimization effect in the low power state.

[0208] The adaptive learning system can dynamically adjust the scheduling and power consumption strategies of future tasks by combining the actual data of historical tasks with the preset resource allocation and power consumption limit. Unlike traditional fixed strategies, this system can continuously optimize according to user behavior patterns, system load and task performance, improving the efficiency of background tasks and battery management capabilities.

[0209] The adaptive learning and optimization mechanism in this step plays a key role in the present invention: by continuously learning user behavior and task execution data, the system can more effectively perform resource scheduling and power management in low-power mode. In particular, in the screen-off state, the system can intelligently adjust the resource usage and preloading strategy of the task to ensure a good user experience while saving power.

[0210] like Figure 2 As shown, in another embodiment of the present invention, a control system for reducing power consumption of a mobile phone after the screen is off is provided, and the system includes:

[0211] The daily behavior data acquisition unit 3011 is used to acquire the user's daily behavior data, including social interaction data and shopping behavior data, and perform preliminary analysis on the behavior data fusion time series;

[0212] A model building unit 3012 is used to build a social behavior model and a shopping behavior model based on the results of the preliminary analysis, and train them respectively to predict the user's behavior trend and purchase intention; wherein the social behavior model and the shopping behavior model use a recursive neural network structure to perform time series modeling;

[0213] Among them, the prediction of the social behavior model is:

[0214]

[0215] in, is the predicted value of social behavior at time t, indicating the predicted social activity; is the social behavior characteristics of the user at time t-1; It is a social behavior fluctuation adjustment item, which is used to handle the situation where users suddenly use social applications frequently or do not use them at all. is a short-term trend term representing social behavior, which is used to capture the rapid changes in user social behavior; α1 and β1 are weight coefficients, representing the impact of the adjustment term on the overall prediction result; W social is the weight matrix obtained through model training; b social is the bias term, which controls the overall translation of the model; f is the activation function, and the nonlinear function ReLU or LeakyReLU is selected to increase the complexity of the prediction;

[0216] The predictions of the shopping behavior model are:

[0217]

[0218] in, is the predicted value of shopping behavior at time t; is the shopping behavior characteristics of the user at time t-1; It is a purchase intention adjustment item, which reflects the short-term changes in users' shopping interests. is the short-term trend term of shopping behavior; α2 and β2 are the corresponding weight parameters, controlling the contribution of the adjustment term; W shop is the weight matrix of the model, which is the core parameter of model learning; λ is the regularization parameter, which is used to limit the complexity of the model and prevent overfitting; g is the activation function, which is used to output nonlinear change results, and LeakyReLU is commonly used; ||W shop || 2 is a regular term used to prevent overfitting of the weight matrix;

[0219] The priority design unit 3013 is used to dynamically set the priority of the background task according to the prediction result; wherein the priority design is as follows:

[0220]

[0221] in, is the priority of the i-th background task, which determines the priority of its resource allocation; is the predicted value of social behavior related to the i-th task; is the predicted value of shopping behavior related to the i-th task; and are the social and shopping power consumption coefficients of the ith task, respectively, reflecting the impact of the task on battery consumption; α1 and α2 are the weight parameters of social and shopping behaviors, indicating the impact of the weights of different types of tasks in the priority calculation; U t is the current remaining power percentage of the system. When the power is low, the task priority will be more strictly regulated; E t is the current load level of the system, which is used to adaptively adjust the task priority; γ is the adjustment coefficient of system load and power, which is used to adjust the trade-off between power and load in different scenarios; It is a behavior fluctuation item, which is used to measure the volatility and suddenness of user behavior in order to adjust the priority;

[0222] Then introduce the resource allocation matrix to allocate specific system resources to background tasks according to their priorities, expressed as:

[0223]

[0224] in, is the resource allocation for the i-th task; Rtotal is the total amount of system resources available; N is the total number of tasks currently running in the background

[0225] Resource allocation unit 3014, used to dynamically allocate resources in combination with priorities to ensure that high-priority tasks obtain sufficient resource support, while controlling resource occupancy of low-priority tasks and optimizing power consumption;

[0226] The low-power loading unit 3015 is used to design a low-power preloading mechanism. When the user turns off the screen, the user preloads data in advance to quickly access certain high-priority tasks, thereby reducing the waiting time after the screen is turned on.

[0227] The system optimization unit 3016 is used to optimize the prediction combination and resource allocation strategy of the social behavior model and the shopping behavior model in combination with the changes in the user's usage habits to adapt to the long-term needs of the user.

[0228] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present invention. In practical applications, technicians in this field can select part or all of them according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0229] In addition, for technical details that are not described in detail in this embodiment, reference can be made to the parameter operation method provided in any embodiment of the present invention, and will not be repeated here.

[0230] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0231] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0232] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory / random access memory, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0233] The above are only preferred embodiments of the present invention, and the scope of the present invention is not limited thereto. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the protection scope of the present invention.

Claims

1. A control method for reducing power consumption of a mobile phone after the screen is turned off, characterized in that: The method comprises: S1. Obtain users’ daily behavior data, including social interaction data and shopping behavior data, and conduct preliminary analysis on the behavior data fusion time series; S2. Based on the results of the preliminary analysis, a social behavior model and a shopping behavior model are constructed, and they are trained separately to predict the user's behavior trend and purchase intention; wherein the social behavior model and the shopping behavior model use a recursive neural network structure for time series modeling; Among them, the prediction of the social behavior model is: in, is the predicted value of social behavior at time t, indicating the predicted social activity; is the social behavior characteristics of the user at time t-1; It is a social behavior fluctuation adjustment item, which is used to handle the situation where users suddenly use social applications frequently or do not use them at all; is a short-term trend term representing social behavior, which is used to capture the rapid changes in user social behavior; α1 and β1 are weight coefficients, representing the impact of the adjustment term on the overall prediction result; W social is the weight matrix obtained through model training; b social is the bias term, which controls the overall translation of the model; f is the activation function, and the nonlinear function ReLU or LeakyReLU is selected to increase the complexity of the prediction; The predictions of the shopping behavior model are: in, is the predicted value of shopping behavior at time t; is the shopping behavior characteristics of the user at time t-1; It is a purchase intention adjustment item, which reflects the short-term changes in users' shopping interests. is the short-term trend term of shopping behavior; α2 and β2 are the corresponding weight parameters, controlling the contribution of the adjustment term; W shop is the weight matrix of the model, which is the core parameter of model learning; λ is the regularization parameter, which is used to limit the complexity of the model and prevent overfitting; g is the activation function, which is used to output nonlinear change results, and LeakyReLU is commonly used; ||W shop || 2 is a regular term used to prevent overfitting of the weight matrix; S3. Dynamically set the priority of the background task according to the prediction result; wherein the priority is designed as follows: in, is the priority of the i-th background task, which determines the priority of its resource allocation; is the predicted value of social behavior related to the i-th task; is the predicted value of shopping behavior related to the i-th task; and are the social and shopping power consumption coefficients of the ith task, respectively, reflecting the impact of the task on battery consumption; α1 and α2 are the weight parameters of social and shopping behaviors, indicating the impact of the weights of different types of tasks in the priority calculation; U t is the current remaining power percentage of the system. When the power is low, the task priority will be more strictly regulated; E t is the current load level of the system, which is used to adaptively adjust the task priority; γ is the adjustment coefficient of system load and power, which is used to adjust the trade-off between power and load in different scenarios; It is a behavior fluctuation item, which is used to measure the volatility and suddenness of user behavior in order to adjust the priority; Then introduce the resource allocation matrix to allocate specific system resources to background tasks according to their priorities, expressed as: in, is the resource allocation for the i-th task; R total is the total amount of system resources available; N is the total number of tasks currently running in the background S4. Dynamically allocate resources based on priority to ensure that high-priority tasks receive sufficient resource support, while controlling resource usage of low-priority tasks and optimizing power consumption; S5. Design a low-power preloading mechanism. When the user turns off the screen, preload data in advance to quickly access certain high-priority tasks, reducing the waiting time after the screen turns on. S6. Based on the changes in user usage habits, optimize the prediction combination and resource allocation strategy of social behavior models and shopping behavior models to adapt to the long-term needs of users.

2. A control method for reducing power consumption of a mobile phone after the screen is turned off according to claim 1, characterized in that: The S1 also includes preprocessing the behavior data; the preliminary analysis of the behavior data fusion time series is specifically: Perform feature mapping on the preprocessed data to quantify the activity of user behavior, including: For social interaction data: in, represents the social activity of the user at time t, w i is the weight of different social data dimensions, u i is the collected user social behavior, n is the total number of social interaction users, and i is the social interaction user i; For shopping behavior data: Among them, w j is the weight of each dimension of shopping behavior, v j is the user's shopping data, m is the total number of shopping users, and j is shopping user j; The processed behavioral characteristics and Combined with the time series, it is represented and stored as follows: Where T is the total number of time segments collected, and The characteristic values ​​of social and shopping behaviors at each moment.

3. A control method for reducing power consumption of a mobile phone after the screen is turned off according to claim 1, characterized in that: The social behavior fluctuation adjustment item The definition is as follows: Among them, μ social is the mean of the social behavior characteristics; N is the size of the backtracking time window, which is used to determine the historical length of the evaluation; The purchase intent adjustment The definition is as follows: in, The user's recent shopping behavior data; c i is the weight of different behavior types, indicating the importance of different shopping behaviors; M is the size of the most recent time window.

4. A control method for reducing power consumption of a mobile phone after the screen is turned off according to claim 1, characterized in that: The behavior fluctuation term Fine-tune the priority based on the rate of change of the user's recent behavior, expressed as: in, is the behavior feature value of the i-th task at time t; M is the length of the retrospective time window, which is used to evaluate the historical period of behavior fluctuations; The social and shopping power consumption coefficients of the i-th task and Among them, the power consumption coefficient of social tasks Reflects the power consumption of the i-th social task and the power consumption coefficient of the shopping task Indicates the power consumption of shopping-related tasks. Dynamically adjust according to its active and passive consumption behavior; wherein the power consumption coefficient of the social task The dynamic adjustment is expressed as: in represents the active consumption coefficient of the i-th task, that is, the power consumption when the task is active; Indicates the passive consumption coefficient of the task, such as the power consumption during data synchronization; ω social A parameter used to adjust the weight between active and passive power consumption.

5. A control method for reducing power consumption of a mobile phone after the screen is turned off according to claim 1, characterized in that: The objective optimization function of S4 is: in, is the amount of resources allocated to the i-th task; is the priority of the ith task, reflecting the resource urgency of the task in the current system; N is the total number of current background tasks; R total is the total resources available in the current system; R battery U is the total amount of resources adjusted based on the battery power, indicating the amount of resources that the system can allocate when the battery power is low; t The current system battery percentage; E t is the current load status of the system, indicating the overall busyness of the system; η is the weight coefficient used to balance the total amount of resources R total and the resource R adjusted based on the battery condition battery ; At the same time, the upper limit of task power consumption is designed, and the power consumption threshold of the task is dynamically set according to the battery status and system load, which is expressed as: in, is the upper limit of power consumption of the i-th task, indicating the maximum power consumption that the task can consume; P max is the maximum power consumption limit of the task; U t The current battery level of the system; E t is the current load status of the system, indicating whether the system is running under high load; power It is the power consumption adjustment factor, which determines the adjustment range of power consumption under different battery and load conditions. When the battery condition is good, the task obtains resources close to the maximum power consumption, and when the battery power is the lowest or the system load is the highest, the power consumption limit will be reduced to the minimum, ensuring that unnecessary tasks do not waste power.

6. A control method for reducing power consumption of a mobile phone after the screen is turned off according to claim 5, characterized in that: The S4 also includes designing a task freezing mechanism. When the system is in an extreme situation, resources are frozen for low-priority tasks to release more resources for high-priority tasks, which is expressed as: in, is the freezing state of the i-th task, 1 means freezing, 0 means not freezing; θ freeze is the priority threshold for task freezing. Tasks below this threshold will be frozen when the battery is low. min The minimum power threshold set for the system. When the power is lower than this threshold, some low-priority tasks need to be frozen.

7. A control method for reducing power consumption of a mobile phone after the screen is turned off according to claim 5, characterized in that: The S5 also includes designing an adaptive adjustment factor, adjusting based on the historical load and power status of the system, and continuously optimizing the resource and power consumption management strategy, which is expressed as: Among them, λ adaptive is a dynamic adjustment factor used to adjust the sensitivity of resource and power management; initial is the initial setting of the adjustment factor value; δ is the adjustment increment coefficient, which controls the adjustment range of the adjustment factor; U t is the battery charge at time t; E t is the system load status at time t; T is the system historical data at the past T times.

8. A control method for reducing power consumption of a mobile phone after the screen is turned off according to claim 5, characterized in that: The S5 specifically includes: Design a low-power preloading mechanism. When the device enters the screen-off mode, although the user will not use the device immediately, the background task still needs to process key operations, which can be expressed as: in, is the preloading coefficient of the i-th task, indicating the preloading amount allowed for the task in low power mode; is the priority of the ith task, which comes from the previous step and is calculated based on the social and shopping behavior prediction and power consumption model; is the resource allocation of the ith task, reflecting the proportion of the task in the system resources; The maximum preloading power consumption threshold set for the system is used to limit the total power consumption of the preloading task to ensure that the system does not consume excessive power due to preloading; is the scheduling coefficient of the i-th task, indicating the resource scheduling priority of the task in the screen-off state; is the preloading coefficient of the i-th task, reflecting the preloading requirement of the task; is the resource allocation ratio of the i-th task; N is the total number of active background tasks in the current system; θ is the balance coefficient, which controls the scheduling ratio of preloaded tasks and actual running tasks; λ preload is the power consumption-preload adjustment factor, which controls the power consumption sensitivity of the control system to the preload task in the low power state; λ base is the initial preload power consumption sensitivity; α is the adjustment increment, which reflects the power consumption control strength of the system on the preload task according to the load and power conditions; U t The remaining battery percentage of the current system; E t It is the current system load status, indicating the overall busyness of the system.

9. A control method for reducing power consumption of a mobile phone after the screen is turned off according to claim 5, characterized in that: The S6 specifically includes: Design an adaptive learning mechanism to learn from the actual resource consumption and power consumption data of tasks through feedback loops, and dynamically optimize task scheduling and power consumption management strategies.

10. A control system for reducing power consumption after a mobile phone screen is turned off, characterized in that: The system comprises: Daily behavior data acquisition unit, used to acquire the user's daily behavior data, including social interaction data and shopping behavior data, and conduct preliminary analysis on the behavior data fusion time series; A model building unit, used to build a social behavior model and a shopping behavior model based on the results of the preliminary analysis, and to train them respectively to predict the user's behavior trend and purchase intention; wherein the social behavior model and the shopping behavior model use a recursive neural network structure for time series modeling; Among them, the prediction of the social behavior model is: in, is the predicted value of social behavior at time t, indicating the predicted social activity; is the social behavior characteristics of the user at time t-1; It is a social behavior fluctuation adjustment item, which is used to handle the situation where users suddenly use social applications frequently or do not use them at all; is a short-term trend term representing social behavior, which is used to capture the rapid changes in user social behavior; α1 and β1 are weight coefficients, representing the impact of the adjustment term on the overall prediction result; W social is the weight matrix obtained through model training; b social is the bias term, which controls the overall translation of the model; f is the activation function, and the nonlinear function ReLU or LeakyReLU is selected to increase the complexity of the prediction; The predictions of the shopping behavior model are: in, is the predicted value of shopping behavior at time t; is the shopping behavior characteristics of the user at time t-1; It is a purchase intention adjustment item, which reflects the short-term changes in users' shopping interests. is the short-term trend term of shopping behavior; α2 and β2 are the corresponding weight parameters, controlling the contribution of the adjustment term; W shop is the weight matrix of the model, which is the core parameter of model learning; λ is the regularization parameter, which is used to limit the complexity of the model and prevent overfitting; g is the activation function, which is used to output nonlinear change results, and LeakyReLU is commonly used; ||W shop || 2 is a regular term used to prevent overfitting of the weight matrix; The priority design unit is used to dynamically set the priority of the background task according to the prediction result; wherein the priority design is as follows: in, is the priority of the i-th background task, which determines the priority of its resource allocation; is the predicted value of social behavior related to the i-th task; is the predicted value of shopping behavior related to the i-th task; and are the social and shopping power consumption coefficients of the ith task, respectively, reflecting the impact of the task on battery consumption; α1 and α2 are the weight parameters of social and shopping behaviors, indicating the impact of the weights of different types of tasks in the priority calculation; U t is the current remaining power percentage of the system. When the power is low, the task priority will be more strictly regulated; E t is the current load level of the system, which is used to adaptively adjust the task priority; γ is the adjustment coefficient of system load and power, which is used to adjust the trade-off between power and load in different scenarios; It is a behavior fluctuation item, which is used to measure the volatility and suddenness of user behavior in order to adjust the priority; Then introduce the resource allocation matrix to allocate specific system resources to background tasks according to their priorities, expressed as: in, is the resource allocation for the i-th task; R total is the total amount of system resources available; N is the total number of tasks currently running in the background Resource allocation unit, used to dynamically allocate resources according to priority, ensuring that high-priority tasks receive sufficient resource support, while controlling resource occupancy of low-priority tasks and optimizing power consumption; The low-energy loading unit is used to design a low-power preloading mechanism. When the user turns off the screen, the preloading data can be used to quickly access certain high-priority tasks, thus reducing the waiting time after the screen turns on. The system optimization unit is used to optimize the prediction combination and resource allocation strategies of the social behavior model and the shopping behavior model based on changes in user usage habits to adapt to the long-term needs of users.

Images